Multiscale Modeling of Brain Dynamics at Criticality

Despite the vast amount of empirical data available on brain structure and function, understanding how large-scale neural dynamics emerge from underlying biological mechanisms remains a fundamental challenge in neuroscience. Experimental observations provide only indirect and partial access to neural processes, leaving key mechanistic questions unresolved, particularly in pathological conditions. This thesis addresses this epistemological gap by developing and integrating multiscale computational models with empirical neurophysiological data, aiming to bridge observable brain dynamics with hidden, patient-specific biological mechanisms. The work is grounded in the view that models are not merely descriptive tools but epistemic instruments that formalize hypotheses, enable causal reasoning, and establish principled mappings between measurable neural signals and controllable biological parameters. Within this framework, both phenomenological and mechanistic models are employed to capture complementary aspects of brain organization and dynamics. At the mesoscopic level, the thesis introduces a novel modular spiking neural network inspired by Hopfield-type attractor model, incorporating biologically plausible spike-timing-dependent plasticity and the ability to store and replay distributed spatiotemporal patterns. This model, on one hand, provides insights into how modularity and temporal coordination support memory capacity and flexible information processing in largescale neural systems, on the other hand, it helps to develop a theoretical analysis of brain criticality, integrating concepts from statistical physics with neural data to analyze empirical data according to the theory of brain criticality. A unifying theoretical perspective throughout the thesis is the concept of brain criticality. By analyzing neuronal avalanches and large-scale spatiotemporal patterns in both empirical recordings and synthetic data, the thesis demonstrates that healthy brain dynamics are best reproduced when models operate near a critical point. Deviations from criticality are systematically linked to pathological conditions, particularly Parkinson’s disease. At the macroscopic level, the work develops and applies neural mass and whole-brain models constrained by empirical structural connectivity. A central contribution is the Dopamine Dynamics Model, so called Dody Model, an original mechanistic large-scale framework that explicitly incorporates dopamine dynamics and connectivity from healthy humans. This model enables the simulation of how neuromodulatory alterations propagate across brain networks and shape collective dynamics. Crucially, the thesis moves beyond forward modelling by introducing a model inversion approach to address the inverse problem inherent in clinical neuroscience: inferring hidden biological mechanisms from observable neural data. Applying this framework to multimodal EEG, and intracranial recordings from Parkinson’s disease patients, both before and after dopaminergic treatment, the work infers patient-specific dopaminergic states directly from empirical signals. This approach enables, for the first time, the virtualization of Parkinsonian patients, allowing in silico exploration of disease mechanisms and therapeutic interventions. Across these components, the thesis demonstrates that multiscale modelling—combined with forward simulation, empirical validation, and model inversion—provides a powerful framework for understanding how microscopic physiological alterations propagate to macroscopic brain dynamics. The results support the hypothesis that healthy human brain activity operates near a critical regime, show how neuromodulation reshapes whole brain dynamics, and illustrate how computational models can bridge the explanatory gap between biological mechanisms, measurable signals, and clinical phenomena. Together, these contributions advance the development of mechanistic digital twins of human brain function and the adoption of the framework of brain criticality, offering new tools for studying neuropathology, with particular emphasis on Parkinson’s disease.